Does PWG order its senses chronologically?
The Poona Dictionary (PD) is built on the OED model and claims historical principles — that its senses run from older to newer. In October 2026 that claim was finally measured (csl-atlas PD sense-order report): PD's senses do tend to rise in date, but only as a tendency (τ = +0.32 on the Dharmamitra clock, a median drift of +265 years from first to last sense, and 40 % of two-sense entries not forward — 26 % run backwards, 14 % tie).
A tendency means nothing without a control. Böhtlingk–Roth (PWG) is that control — a dictionary of the same century with no OED remit — and the answer it gives is the reason this page exists: PWG orders its senses more chronologically than PD does. So PD's sense order is not distinctive, and "historical principles" is not evidenced by the arrangement of senses.
This page is about the order of senses inside an entry, not the order of quotations inside a sense. The latter is a different measurement, on a different unit — see the PD citation-order report.
What was measured
For PWG, PW, MW and AP: each numbered sense division contributes the earliest datable citation filed under it; the sequence of those dates in sense order is then tested for a rising tendency. Two date instruments are used, and both were already the ones behind the PD figure, so the comparison is like-for-like:
- DCS — the Digital Corpus of Sanskrit chronology, matched at Dice ≥ 0.90 to the work a citation abbreviation names.
- Dharmamitra — its post-median date for the work, earliest stratum of a base title, matched at Dice ≥ 0.82.
Citation abbreviations are resolved from each dictionary's own key
(csl-guides abbreviations), longest-prefix; AP
borrows the documented Apte (AP90) key, and MW resolves through the vendored
mw-source-abbreviations.tsv. Thresholds are fixed and were never relaxed per dictionary.
The statistic is the one PD's own report uses, imported rather than re-implemented: the two-sense forward rate (later sense, later date) with an exact Clopper–Pearson 95 % interval, Kendall τ between sense position and earliest date for entries with at least three datable senses, a 5,000-replicate within-entry permutation null, and the median first→last drift.
The numbers
τ is on the Dharmamitra clock (the clock the PD figure quotes); the DCS column replicates.
"Resolved" is the share of the dictionary's <ls> citations that reached a dated work at all —
the honest ceiling on what this instrument can see.
| Dictionary | Entries ≥3 | τ (dm) | τ (dcs) | Two-sense forward rate (95 % CI) | Two-sense entries | Median drift | <ls> resolved |
|---|---|---|---|---|---|---|---|
| PD (Poona) | 706 | 0.323 | — | 60.1 % * | 582 | 265 y | — |
| PWG (Böhtlingk–Roth) | 3,343 | 0.406 | 0.395 | 54.8 % (53.5–56.0) | 6,273 | 184 y | 67.1 % |
| MW (Monier-Williams) | 3,882 | 0.227 | 0.256 | 49.7 % (48.2–51.2) | 4,367 | 10 y | 96.8 % |
| PW (Böhtlingk shorter) | 281 | -0.041 | 0.089 | 37.3 % (34.1–40.6) | 895 | 0 y | 51.5 % |
| AP (Apte) | 1,622 | -0.042 | -0.147 | 36.9 % (34.8–39.0) | 2,042 | 0 y | 83.5 % |
- PD's rate is quoted from the published PD summary, which reports it without an interval; every control row carries its exact 95 % CI.
PWG's tendency is the strongest of the five, and it holds in both size bands — τ = 0.421 on the 3,037 entries with 3–7 senses and 0.258 on the 306 with eight or more — so the giant root entries (gam carries 221 numbered Bedeutungen) are not carrying the result.
The verdict
The comparison rule was fixed before the control numbers were computed, and the vocabulary is its own: the claim stands only if PD beats every scored dictionary by τ ≥ 0.10 with a Holm-adjusted one-sided permutation p ≤ 0.05; otherwise it is withdrawn.
| Dictionary | τ | Δτ = τPD − τ | 95 % CI | one-sided p | Holm p | clears the floor |
|---|---|---|---|---|---|---|
| PWG | 0.406 | -0.082 | -0.125 … -0.040 | 1.0000 | 1.0000 | no |
| MW | 0.227 | 0.096 | 0.054 … 0.138 | 0.0002 | 0.0008 | no |
| PW | -0.041 | 0.364 | 0.290 … 0.440 | 0.0002 | 0.0008 | yes |
| AP | -0.042 | 0.365 | 0.316 … 0.414 | 0.0002 | 0.0008 | yes |
Outcome: WITHDRAWN. PWG's τ of 0.41 sits above PD's 0.32 — a reversal, not a tie, and its bootstrap interval excludes zero. MW, at 0.23, is the close call: it falls 0.004 short of the floor and would not decide the verdict on its own. PW and AP show no chronological order at all. The reading that PD's sense arrangement is historically distinctive is therefore withdrawn; the tendency is shared with dictionaries that do not carry the OED's programme.
A worked entry: guru
PWG numbers a sense (1〉) and then subdivides it with lettered sub-senses (a〉, — b〉). The
measurement takes the numbered sense as the unit — a lettered sub-sense folds into its
parent, so the parent's date is the earliest among all its sub-senses, exactly as a nested
division behaves in PD. Trimmed from csl-orig/v02/pwg/pwg.txt, record L 22685 (vol. 2,
p. 767):
<L>22685<pc>2-0767<k1>guru<k2>guru/
<div n="1"> 1〉 adj. f. {#gurvI#}; compar. {#ga/rIyaMs#}
<ls>P. 6,4,157</ls>. <ls>VOP. 7,56</ls>. …
<div n="2"> a〉 {%schwer%} (<ab>Gegens.</ab> {#laGu#}) …
<ls>ṚV. 1,39,3</ls>. <ls>AV. 9,3,24</ls>. …
<div n="2">— b〉 {%gross, ausgedehnt%}
(dem äussern Umfange nach), = {#mahant#} … <ls>PAÑCAT. II, 199</ls>. …
(Trimmed and wrapped for display; the record's own line breaks are different.)
| Sense | What it holds | Earliest dated citation (Dharmamitra) |
|---|---|---|
1〉 adj. — heavy, great, long (with sub-senses a–e) | Ṛgvedic, with the classical grammarians alongside | −1125 (Ṛg-Veda) |
2〉 m. — a venerable person; father, mother, elder relative | classical | −369 |
3〉 f. — pregnant woman; a teacher's wife | lexicographers only, no datable work | — (skipped) |
The entry rises as the tendency predicts, and it also shows the two honest limits: the third sense has no dated citation and is skipped, not imputed, and the earliest citation is the earliest quotation filed under the sense, not a date Böhtlingk recorded. (Both figures are from the 19-10-2026 run; the sense-level dates are not stored as data on this site.)
What this lets you say — and not say
- PWG's arrangement is meaning-driven first and historical second: 54.8 % of two-sense entries run forward, so almost half do not. Treat sense position as a weak prior on age, never as a date.
- The forced canonical order (literal → figurative → technical) is shared by every dictionary and is exactly what the control measures; it is reported, never corrected away.
- The resolution ceiling is the binding constraint: 67.1 %
of PWG's 720,203
<ls>citations reached a dated work, because PWG quotes lexicographers, grammarians and medieval compendia that the two chronologies do not date. - DCS and Dharmamitra are one lineage, not two instruments (ρ = 0.94 on 56 shared works), so the two clocks replicate a chronology rather than doubling the evidence.
Reproduce
The measurement lives in csl-atlas, not here:
scripts/pd_sense_order_compare.py(--build,--check,--render) and its packagescripts/pd_compare/— the PD row is recomputed from the committed TSV and must reproduce the PD summary exactly before any control is read.data/pd/pd_sense_order_compare_summary.json— every number on this page, with its denominators.- Report · nine pinned forensic tests that fail if the verdict flips silently.
This page renders src/data/pwg-sense-order.json, a projection of that summary generated by
scripts/build-pwg-sense-order.mjs
(npm run build:pwg-sense-order, --check to verify). The source SHA-256 is recorded in the JSON,
so a re-run against a drifted artefact fails instead of re-vendoring silently.
See also
- Böhtlingk–Roth (PWG) — the dictionary's own page
- How to read PWG — the dictionary-literacy pack
- The citation graph — what PWG quotes, and how often
- Corpus attestation — the same works in the dated corpus